Amjad Masad's 8B LLM chess engine hits ~1500 Elo and beats GPT-5.6 with high reasoning

X/Twittervia @amasad·low signalXBlueskyLinkedInCopy link

Replit's CEO reported that his LLM chess engine reached roughly 1500 Elo and 'consistently beats frontier models and Stockfish level 0,' describing it as an 8B model outperforming GPT-5.6 with high reasoning and response chaining. Single-source and self-reported with no published eval harness, so treat the Elo figure as a claim rather than a measurement. The interesting angle for builders is the argument that a small specialised model plus scaffolding beats a frontier model on a narrow verifiable task.